Adaptive Segmentation in Speech Coding for Packet Loss Channels

The scope of this research project is packet-based transmission of speech over IP networks i.e. Voice over IP (VoIP). This application gives rise to some special requirements to the speech coding algorithm. Besides obtaining a high speech quality at a low bit rate, the possibility of packet loss suggests that packets should be self-contained, so that each frame can be decoded without knowledge of past or future packets. In conventional speech coding, the speech signal is segmented into frames of fixed duration (typically 20-30 ms.), and a fixed number of bits assigned to the representation of each of these. However, this approach is far from optimal. Instead, we designed a coder based on the rate-distortion principle, where both segmentation and bit-allocation are signal adaptive. Objective and subjective results show that the proposed methods can significantly improve the speech signal quality without increasing the bit rate. (Christoffer A. Rødbro, Søren Holdt Jensen, Jesper Jensen (Delft University of Technology), Richard Heusdens (Delft University of Technology))